Accept applied/reverted/failed in device status reports

internal/store/store.go

@@ -7,6 +7,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"time"
 
 	"github.com/jackc/pgx/v5"
 	"github.com/jackc/pgx/v5/pgxpool"
@@ -279,11 +280,16 @@ func (s *Store) UpsertDevice(ctx context.Context, d model.Device) error {
 	})
 }
 
-// RecordDeviceStatus stores the generation an agent reports as applied.
-func (s *Store) RecordDeviceStatus(ctx context.Context, name string, generation int64) error {
-	tag, err := s.pool.Exec(ctx,
-		`UPDATE devices SET reported_generation = $2, last_seen = now() WHERE name = $1`,
-		name, generation)
+// RecordDeviceStatus stores an agent's apply report. reported_generation only
+// advances on applied; every report updates the last-status columns.
+func (s *Store) RecordDeviceStatus(ctx context.Context, name string, r model.StatusReport) error {
+	tag, err := s.pool.Exec(ctx, `
+		UPDATE devices SET
+			reported_generation = CASE WHEN $2 = 'applied' THEN $3 ELSE reported_generation END,
+			status = $2, status_generation = $3, status_error = $4,
+			status_at = now(), last_seen = now()
+		WHERE name = $1`,
+		name, string(r.Status), r.Generation, r.Error)
 	if err != nil {
 		return err
 	}
@@ -296,10 +302,14 @@ func (s *Store) RecordDeviceStatus(ctx context.Context, name string, generation
 func (s *Store) GetDevice(ctx context.Context, name string) (model.Device, error) {
 	var d model.Device
 	var resolver, settings, reachable []byte
+	var st model.AgentStatus
+	var statusAt *time.Time
 	err := s.pool.QueryRow(ctx,
-		`SELECT name, class, COALESCE(fabric, ''), resolver, settings, reachable_prefixes
+		`SELECT name, class, COALESCE(fabric, ''), resolver, settings, reachable_prefixes,
+		        reported_generation, status, status_generation, status_error, status_at
 		 FROM devices WHERE name = $1`, name,
-	).Scan(&d.Name, &d.Class, &d.Fabric, &resolver, &settings, &reachable)
+	).Scan(&d.Name, &d.Class, &d.Fabric, &resolver, &settings, &reachable,
+		&st.AppliedGeneration, &st.Status, &st.Generation, &st.Error, &statusAt)
 	if errors.Is(err, pgx.ErrNoRows) {
 		return d, ErrNotFound
 	}
@@ -312,6 +322,10 @@ func (s *Store) GetDevice(ctx context.Context, name string) (model.Device, error
 	if err := json.Unmarshal(reachable, &d.ReachablePrefixes); err != nil {
 		return d, err
 	}
+	if statusAt != nil {
+		st.ReportedAt = *statusAt
+		d.AgentStatus = &st
+	}
 	return d, json.Unmarshal(settings, &d.Settings)
 }
 

internal/store/store_integration_test.go

@@ -149,3 +149,57 @@ func TestNATTierRoundTrip(t *testing.T) {
 		t.Errorf("expected nat rows to cascade-delete with device, got %d", len(nats))
 	}
 }
+
+func TestRecordDeviceStatus(t *testing.T) {
+	s := newTestStore(t)
+	ctx := context.Background()
+	if err := s.UpsertDevice(ctx, model.Device{Name: "fw1", Class: model.ClassFirewall}); err != nil {
+		t.Fatalf("upsert device: %v", err)
+	}
+	if d, _ := s.GetDevice(ctx, "fw1"); d.AgentStatus != nil {
+		t.Fatalf("unreported device should have no agent_status: %+v", d.AgentStatus)
+	}
+
+	report := func(r model.StatusReport) model.AgentStatus {
+		t.Helper()
+		if err := r.Normalize(); err != nil {
+			t.Fatalf("normalize: %v", err)
+		}
+		if err := s.RecordDeviceStatus(ctx, "fw1", r); err != nil {
+			t.Fatalf("record %+v: %v", r, err)
+		}
+		d, err := s.GetDevice(ctx, "fw1")
+		if err != nil || d.AgentStatus == nil {
+			t.Fatalf("get device: %v %+v", err, d.AgentStatus)
+		}
+		return *d.AgentStatus
+	}
+
+	// Old agent payload: generation only.
+	got := report(model.StatusReport{Generation: 5})
+	if got.AppliedGeneration != 5 || got.Status != model.StatusApplied || got.Generation != 5 || got.ReportedAt.IsZero() {
+		t.Errorf("old payload: %+v", got)
+	}
+
+	got = report(model.StatusReport{Status: model.StatusApplied, Generation: 6})
+	if got.AppliedGeneration != 6 || got.Status != model.StatusApplied {
+		t.Errorf("applied: %+v", got)
+	}
+
+	got = report(model.StatusReport{Status: model.StatusReverted, Generation: 7, Error: "api unreachable"})
+	if got.AppliedGeneration != 6 || got.Status != model.StatusReverted || got.Generation != 7 || got.Error != "api unreachable" {
+		t.Errorf("reverted must not advance applied generation: %+v", got)
+	}
+
+	got = report(model.StatusReport{Status: model.StatusFailed, Generation: 8, Error: "nft: syntax error"})
+	if got.AppliedGeneration != 6 || got.Status != model.StatusFailed || got.Generation != 8 || got.Error != "nft: syntax error" {
+		t.Errorf("failed must not advance applied generation: %+v", got)
+	}
+
+	if err := s.RecordDeviceStatus(ctx, "fw1", model.StatusReport{Status: "bogus", Generation: 9}); err == nil {
+		t.Error("db should reject unknown status")
+	}
+	if err := s.RecordDeviceStatus(ctx, "nope", model.StatusReport{Status: model.StatusApplied}); err != store.ErrNotFound {
+		t.Errorf("unknown device: want ErrNotFound, got %v", err)
+	}
+}